Deep Learning-Based Remaining Useful Life Estimation of Bearings with Time-Frequency Information.

Area of Science:

• Mechanical Engineering
• Artificial Intelligence
• Signal Processing

Background:

• Accurate prediction of bearing remaining useful life (RUL) is crucial for industrial system safety and stability.
• Traditional RUL prediction methods often rely on complex physical modeling and struggle with intricate systems.
• Existing approaches may lack interpretability, hindering trust and understanding of the prediction process.

Purpose of the Study:

• To propose an end-to-end deep learning method for remaining useful life (RUL) prediction of bearings.
• To enhance the accuracy and reliability of RUL predictions in industrial applications.
• To improve the interpretability of the RUL prediction model.

Main Methods:

• Utilized Short-Time Fourier Transform (STFT) for signal preprocessing.
• Developed a Convolutional Neural Network (CNN) integrated with a Long Short-Term Memory (LSTM) network to capture temporal dependencies.
• Incorporated a Convolutional Block Attention Module (CBAM) to enhance feature extraction and model interpretability.

Main Results:

• The proposed method demonstrated superior performance on the 2012PHM dataset compared to existing techniques.
• The integration of STFT, CNN, LSTM, and CBAM effectively predicted bearing RUL.
• The model provided insights into its decision-making process, enhancing interpretability.

Conclusions:

• The developed end-to-end deep learning approach offers an effective solution for bearing RUL prediction.
• The method shows significant potential for improving the safety and stability of industrial systems.
• The enhanced interpretability of the model facilitates better understanding and trust in its predictions.